Author's personal copy

Transcription

1 DOI /s y ORIGINAL PAPER IN PHILOSOPHY OF SCIENCE Structural realism and the nature of structure Jonas R. Becker Arenhart Otávio Bueno Received: 28 November 2013 / Accepted: 28 September 2014 SpringerScience+BusinessMediaDordrecht2014 Abstract Ontic Structural Realism is a version of realism about science according to which by positing the existence of structures, understood as basic components of reality, one can resolve central difficulties faced by standard versions of scientific realism. Structures are invoked to respond to two important challenges: one posed by the pessimist meta-induction and the other by the underdetermination of metaphysics by physics, which arises in non-relativistic quantum mechanics. We argue that difficulties in the proper understanding of what a structure is undermines the realist component of the view. Given the difficulties, either realism should be dropped or additional metaphysical components not fully endorsed by science should be incorporated. Keywords Structural realism Structure Underdetermination Realism 1 Introduction Ontic Structural Realism (OSR) is one of the most promising ways to develop a form of realism in contemporary philosophy of science. It advances a metaphysical thesis that aims to overcome two of the main difficulties for the realist: the problem of securing reference and approximate truth through theory change the target of the so-called pessimist meta-induction and the problem of metaphysical underdetermination the fact that the metaphysical nature of the objects posited by certain J. R. Becker Arenhart Department of Philosophy, Federal University of Santa Catarina, Florianópolis, SC , Brazil jonas.becker2@gmail.com O. Bueno ( ) Department of Philosophy, University of Miami, Coral Gables, FL 33124, USA otaviobueno@mac.com

2 scientific theories is underdetermined by such theories. To solve these difficulties, the ontic structural realist advances a metaphysical thesis to the effect that structures and relations are the fundamental components of the world; objects are secondary they should either be eliminated or at best re-conceptualized in structural terms (see Ladyman 1998, and French and Ladyman 2003, 2011). How can the appeal to an ontology of structures save realism given the pessimist meta-induction? Recall that according to the pessimist meta-induction, what in the past were taken to be our best scientific theories are now recognized as being defective; terms that were thought of as having reference in fact do not refer, and theories that were thought of as being true (or approximately true) are now recognized as being false. Similarly, the argument goes, our current best theories will probably have the same fate sooner or later it is likely that they will also be shown to be false. Thus, it is unclear that one should believe that these theories are true (or approximately true) and that their terms refer. An ontology of structures overcomes this difficulty by allowing for changesin the objects that are referred to in theory change, but insisting that a common structure is preserved through scientific revolutions. That is, in the dynamics of theory change, although the objects referred to by the relevant theories may change, there is structural continuity through the coming and going of the theories in question. In the end, we should be realist about structure, not about the posited unobservable objects. How can the appeal to an ontology of structures save realism given the metaphysical underdetermination? To address this second main motivation for OSR, let us turn briefly to a dispute about the metaphysics of non-relativistic quantum mechanics (see French and Krause 2006, especially Chapter 4). A central issue to be considered is the metaphysical nature of quantum particles. Two options emerge in this context: particles as individuals (according to which, roughly, particles have well-defined identity conditions, can be identified and re-identified); particles as non-individuals (according to which, roughly, identity is not well defined for quantum particles, there are no identity conditions for them). These options are, of course, object-oriented ontologies (in a broad sense of object that does not require well-defined identity conditions for something to be an object). The main problem for such ontologies in quantum mechanics concerns the fact that the theory, by itself, is unable to determine which option is the right one. So, the argument goes, as far as quantum mechanics is concerned, both ontological options are equally acceptable. According to the proponents of OSR this situation is untenable for a scientific realist: realists should be able to determine the nature of the entities they are realist about (see, for instance, Ladyman 1998, p. 420). Since it is unclear how to do that, given metaphysical underdetermination, one is better off avoiding objects altogether particularly those whose metaphysical status cannot be determined keeping commitment only to the structure that is common to both options (French and Ladyman 2003, p.37).by restricting the commitment only to structure, one can ensure that one s ontology does not overstep what is sanctioned by the sciences. In both motivations for OSR, the same metaphysical component plays the decisive role: structure is posited as that about which one is realist. In the first case,

3 structure provides the common basis across theory change to anchor one s realism. In the second case, structure allows one to preserve realism in face of metaphysically contentious objects, by providing a common basis among rival metaphysical views regarding the nature of the relevant particles. As a result, one can then resist sliding into anti-realism. For this reason, if there were an adequate account of what a structure is clearly, a fundamental requisite to make sense of OSR this kind of realism would be in a privileged situation: it would be able to solve the problems that challenged earlier forms of scientific realism while being clearly and intelligibly formulated. Before we proceed, we should note that there is a plethora of positions under the heading of scientific structuralism, and the same goes for the ontic brand of this family of views. Our focus, in this paper, is on versions of OSR that conceive of objects as either eliminable (a position associated mainly with Steven French) or as ontologically derivative from relations and the structure of which they are a part (a view defended by James Ladyman; see French and Ladyman 2011). Unless otherwise stated, when we write structural realism, we mean ontic structural realism of those two specific sorts. This means that versions of OSR allowing for objects as primary entities on an equal footing with relations, such as Moderate Structural Realism (MSR), and other variants that allow for objects as primary entities are not our main target. We aim to examine them explicitly in a future work. 1 Our aim in this paper is to show that it is unclear that a proper characterization of structure suitable for ontic structural realism can be offered. We argue that there are far too many distinct ways of characterizing structure and relations, and as a result, the combination of realism and a metaphysics of structures becomes, at best, problematic and, at worst, incoherent. We begin, in Section 2, by presenting arguments from a formal point of view. The nature of structures and the representational apparatuses used to characterize them are critically examined. In Section 3, we address the problem of the metaphysical nature of structures and relations. In particular, the ambiguous status of such metaphysical nature is emphasized. We conclude with a discussion of the tenability of combining realism and structuralism. In light of the difficulties of the position, something must go, and the obvious candidate, if we are to keep structures, we argue, seems to be realism. 2 Structure and mathematics What are the prospects for realism about structures? Within structural realism, we noted, structures play a key role in solving difficulties of traditional realism. Thus, positing such structures may seem warranted. But just what is structure? Of course, this question has been raised before. We argue, though, that no matter how it is answered, problems will emerge for the ontic structural realist. In this section, we examine the question in the context of various mathematical representational apparatuses for structures. We divide the section into two parts. In the first, we argue that 1 AclassificationofdistinctversionsofOSRispresentedinAinsworth(2010).

4 defenders of OSR are ultimately unable to avoid commitment to objects. In the second, we argue that OSR is unable to identify the structure of the world given the diversity of candidates to get the job done. 2.1 Mathematical frameworks and commitment to objects The adoption of OSR involves two conflicting features, which bring a tension to those who intend to provide a structural realist account of the metaphysics of structures. On the one hand, ontic structural realists argue that theories are better characterized in accordance with the semantic approach, rather than in terms of the syntactic view of theories and related approaches to structure based on Ramsey sentences. In particular, within the semantic tradition, the partial structures approach has been employed to accommodate both vertical relations between scientific theories and data, and horizontal relations among distinct theories (Ladyman 1998; French and Ladyman 2011;daCostaandFrench2003). 2 On the other hand, the semantic approach is typically formulated in terms of set-theoretic structures. 3 But this commitment to set theory, we argue, introduces objects as key components in the characterization of structures, and is responsible for the tension. As a framework to define what a structure is, set theory has at least two clear advantages: conceptual clarity and familiarity. It is well known what set-theoretic structures are and how they are constructed: they can be characterized as ordered pairs E = D, R consisting of a domain of objects and a family of relations among those objects, all of which are found in the set-theoretic hierarchy (see da Costa and Rodrigues 2007 for a general theory of structures). Relations are then defined in terms of the objects that belong to the domain, and not the other way around. Given a structure, the existence of relations, as particular sets, depends on the existence of the elements of the domain: without the objects in D there would be no relations, and, hence, no structure in the set-theoretic sense. This is part and parcel of the iterative conception of set, according to which sets are constructed in stages, and are determined by their elements. Thus, objects are basic in set theory: either sets themselves are objects, such as the empty set in pure set theory and the sets formed from it, or in impure set theory, objects that are not sets the Urelemente are used to form additional sets, in which case the Urelemente are also basic. However, for the reasons discussed above, objects are not allowed as primary entities in ontic structural realism. So, if the structural realist s characterization of structures is implemented in terms of set theory, some maneuver needs to be adopted to defuse the resulting commitment to objects. To overcome this difficulty one can maintain that structures should be read and understood from right to left, from the relations to the objects. This would allow for objects to be somehow constituted by, or at least re-conceptualized via, the rela- 2 For a succinct discussion of partial structures and their application in the philosophy of science, see Bueno and da Costa (2007). 3 Landry (2007)alsohighlightstheintimateconnectionbetweenthesemanticviewandsettheory,although her concerns are different from ours.

5 tions (French and Ladyman 2003, and also French 2010). This strategy is called the Poincaré manoeuvre bystevenfrench(2012, p. 23). According to it, objects are used merely as heuristic devices or stepping stones to obtain the structure. After the structure is characterized, the objects are left behind: either they are taken as metaphysically irrelevant entities or are only conceived as being derived from the relations, depending on the kind of OSR that is assumed. The central point is to ensure that objects are, at best, obtained after the relations are given and obtained from them, not the other way around. Given this maneuver, the need for using objects in set theory to characterize structures poses no threat to a structure-oriented metaphysics. In the end, it is ultimately a matter of knowing how to read the structure, and to realize that any reference to objects to begin with is purely heuristic. This maneuver, however, faces significant difficulties. First, in set theory, structures are obtained as elements of the set-theoretic hierarchy. As noted, on the set-theoretic account of structure, objects are used to construct relations and structures, not vice versa (see, in particular, the theory of structures in da Costa and Rodrigues 2007). The following argument then emerges: (i) Realists about the structure of theories must be realist about the mathematical parts of these theories, since it is not possible to separate their mathematical content from their nominalistic content (see Azzouni 2011). The mathematical content refers tomathematicalobjects,relations and functions; the nominalistic content does not. Furthermore, (ii) if set theory is used to characterizethemathematicalstructures in question, sets as abstract entities will thereby be included among the structural realist s commitments. Thus, a commitment to objects sets and their members emerges in the structural realist s metaphysics right from the start. Let s call this argument the commitment-to-objects argument. 4 This argument has two important assumptions: (a) It depends on the impossibility of separating the nominalistic and the mathematical content of scientific theories. (b) It also depends on the use of set theory inthecharacterizationofmathematical structures. Let s discuss each of these assumptions in turn. (a) It is now widely acknowledged that the major attempt at providing a demarcation between the nominalistic content and the mathematical content of a scientific theory Hartry Field s nominalistic program (see Field 1980) has not succeeded at establishing the intended result(forasurveyandreferences, see Bueno 2013). And it is unclear which additional resources are available to implement such a demarcation (see Azzouni 2011 for further discussion). 4 Note that we are not invoking the indispensability argument here, as will become clear below. Our point is that by using set theory, the structural realist is thereby committed to objects unless a proper nominalization of set theory itself is developed. (But, we will also argue, such a nominalization may conflict with the realist component of structural realism.) Note also that the point goes through independently of how much set theory is ultimately used. So it doesn t matter whether one is dealing with a highly mathematized science or with a less mathematized one. As long as set theory is used by the structural realist (absent a full nominalization of that theory), a commitment to objects emerges.

6 Thus, the assumption regarding the impossibility of separating the nominalistic and the mathematical content of scientific theories is one that is reasonable to invoke. Note, however, that the commitment-to-objects argument is neutral on a stronger claim: the indispensability of mathematics. The claim that scientific theories cannot be formulated without quantification over mathematical objects, relations and functions which would make these objects, thereby, indispensable to such theories is not presupposed in the argument. The argument s premises and conclusion are certainly compatible with mathematics being indispensable, but the indispensability is not required for the argument to go through. Let s see why this is the case. As is well known, the indispensability argument aims to establish commitment to objects that are indispensable to our best theories of the world (for discussion and references, see Colyvan 2001). It was originally designed by W. V. Quine (see, e.g., 1960) to force those who are realist about scientific theories to become realist about the mathematics that is indispensably used in such theories. In fact, the argument is supposed to conclude that the grounds that are invoked to establish ontological commitment in science are the same that establish commitment to those mathematical objects and structures that are indispensable to the relevant scientific theories. But the commitment-to-objects argument does not rely on such indispensability. After all, the structural realist s commitment to the mathematical content of scientific theories emerges from the inseparability of that content from the nominalistic content of scientific theories, and from the fact that, given realism about the physical world, the structural realist is committed to the nominalistic content which is, as noted, the content that refers to the non-mathematical features of the world. The commitment to the mathematical content then follows independently of indispensability considerations. One may argue that the inseparability of the mathematical content and the nominalistic content of a scientific theory just is what the indispensability of mathematics amounts to. But this is not right. We understand the indispensability thesis as the claim that scientific theories cannot be formulated without reference to mathematical objects, relations and functions. We understand the inseparability thesis as the claim that it is not possible to separate the nominalistic content and the mathematical content of a scientific theory. The indispensability thesis may entail the inseparability thesis, but not the other way around. After all, from the fact that the nominalistic content and the mathematical content of a scientific theory cannot be separated, it does not follow that reference to mathematical objects, relations and functions is indispensable. For a different formulation of the relevant scientific theory can be provided in terms of a different framework in which no reference to such mathematical objects, relations and functions is found. For example, instead of using set theory as the underlying mathematical framework, one can use second-order mereology plus plural quantification (see Lewis 1991, 1993). This framework is committed to mereological atoms (admittedly, a lot of them!), but not to sets. As Lewis shows, as long as there are inaccessibly many mereological atoms, one can mimic the expressive resources of set theory without thereby having the same commitments that set theory

7 has. 5 The important feature is that the commitment-to-objects argument only requires the inseparability thesis, not the indispensability one. Motivated by these considerations, perhaps the structural realist could try to resist the commitment-to-objects argument by adopting an anti-realist view about mathematics while preserving realism about science. More specifically, maybe the structural realist could adopt a deflationary nominalist view about mathematics (such as the one developed and defended by Azzouni 2004; for some discussion, see Bueno 2013). The deflationary nominalist grants that mathematics is indeed indispensable to science, but resists the conclusion that this provides any reason to be committed to the existence of mathematical objects and structures. This is achieved by distinguishing quantifier commitment (the mere quantification over the objects of a given domain, independently of their existence) and ontological commitment (the quantification that commits one ontologically to the existence of something). If the quantifiers are not interpreted as being ontologically loaded, the fact that one quantifies over certain objects or structures does not entail that such objects or structures exist. It just means that the relevant objects or structures are talked about, that they are objects of thought, as it were. Thus, the structural realist, despite quantifying over set-theoretic structures, need not be committed to their existence, nor to any claim that these structures fully capture the nature of the structures one should be realist about. The problem with the introduction of ontologically neutral quantifiers in the context of structural realism is that, given these quantifiers, it is unclear how structural realists will manage to specify what their realism amounts to. Unless they provide an independent mechanism of access to, and specification of, the structures they are realist about, the use of ontologically neutral quantifiers will ultimately remove all ontological content from structural realism. It is now left entirely unspecified what, exactly, they are supposed to be realist about. In this way, realism about the physical world seems to have been lost. Perhaps structural realists could insist that the structures they are realist about are those that were obtained via inference to the best explanation as part of the success of science. Mathematical structures only represent the nominalistic (physical) content, which is the content structural realists are ultimately committed to; they need not be committed to the mathematical content. In other words, the set theory that structural realists invoke only play a representational role; it does not provide any guide to the commitments structural realists have. However, with this response, the initial problemsimply returns: Howcan the nominalistic content be specified without a proper nominalization of mathematics in the first place? If quantifiers are not ontologically neutral, given the use of set theory structural realists are committed to objects (namely, sets), which is incompatible with 5 One may worry about the full success of Lewis construction. Since the notion of inaccessibility is fundamentally set-theoretic in nature, aren t sets still presupposed (Bueno 2010)? Even if the proposed reconstruction is expressively equivalent to set theory, is it in fact as effective for the formulation of empirical theories as set theory is? These are fair concerns, but they are also beside the point in this context. The purpose of the Lewis example is just to make a conceptual point, namely, that the inseparability and the indispensability theses are not the same. We need not argue that the indispensability thesis is in fact false; only that it can be.

8 their insistence that structures, rather than objects, are fundamental. Alternatively, if quantifiers are ontologically neutral, it is unclear how structural realists can specify what they are realist about, since such quantifiers will removeall ontological commitment from what is quantified over even if one quantifies over what was obtained, by means of inference to the best explanation, on the basis of the success of science. Perhaps the structural realist could maintain that true existential statements that follow from our best theories indicate such ontological commitment. But with ontologically neutral quantifiers in place, this suggestion would not be enough to express ontological commitment, since these quantifiers only indicate that some part of the domain is being considered, not that what is being quantified over exists. An existence predicate needs to be introduced for that. But what should the content of this predicate be? One possibility is to propose that the existence predicate expresses ontological independence: those things that are ontologically independent from our linguistic practices and psychological processes exist (Azzouni 2004). There is, however, significant disagreement in discussions of realism in science about what kinds of things are (or are not) ontologically independent from us. Standard scientific realists who are committed to the existence of quantum particles insist that these particles are ontologically independent from us. Ontic structural realists resist that commitment, insisting that ontological commitment to things of such dubious metaphysical status should be avoided. If these realists about science are also platonist about mathematics in particular, about mathematics used in science they will insist that mathematical structures exist, given that these structures are ontologically independent from us. In contrast, if these realists are nominalist about mathematics, they will point out that, since mathematical structures are not ontologically independent from us we made them up, after all these structures do not exist. It is, thus, unclear that ontological independence is of much use in such ontological debates. But perhaps the structural realist may respond by noting that the appropriate existence predicate should identify a suitable mechanism of detection of the relevant structures. After all, it is only with such a detection mechanism that the relevant mathematical structures (suitably interpreted)can haveany empirical significance. If, however, there is such a detection mechanism, the burden is now on structural realists to describe it, show how it functions, and specify precisely how such mechanism yields a stable account of the nature of the structures they should be realist about. It is only after this is done that their view would secure the relevant realist content. But the difficulty is to ensure that the usual mechanisms of detection (such as various scientific instruments used in scientific practice) detect structures rather than particular objects. Consider the micrograph from an electron microscope. It may be argued that on the surface of that image we find the representation of particular objects: whatever objects that were present in the sample when the micrograph was generated. Rather than a commitment to structures, on this view, micrographs provide information about the relevant objects. The worry is that structural realists may end up presupposing objects as part of the specification of whatever detection mechanisms they invoke. In response, structural realists could argue that micrographs do exhibit structural features: the various relations among the objects that are represented in the image.

9 Moreover, they continue, those structural features correspond to structural components of the world. But it is unclear that this response is really open to structural realists. Micrographs can certainly display structural traits, but how can structural realists make sense of these traits if they are formulated in terms of relations among objects in the sample? As an illustration, consider a micrograph produced by a transmission electron microscope, which represents ribosomes in a cell. The micrograph represents the ribosomes as located in a particular region of the cell, say, near the membrane. It also represents them as bearing some spatial relations to other ribosomes and other cellular components. We can grant that these features are structural: they display relations among objects, after all. However, in order for the features to be structural, ribosomes need to be taken as objects rather than structures: a structural understanding of ribosomes is obtained via the relations they bear to other cellular components. But this means that ribosomes, as the terms in the various relations, are ultimately understoodasobjects. As aresult, objects are ultimately presupposed, and we end up with an approach that ontic structural realists are unable to embrace. The advocate of ontic structural realism may respond by arguing that, for conceptual considerations, researchers may need to introduce objects, which bear a variety of relations, at certain stages of their inquiry in a particular field. The ribosome case is not different. However, once ribosomes are properly considered, they are best understood as involving a plurality of relations that hold between items provisionally postulated as objects, that is, as relation-bearing items. However, this means that ribosomes are ultimately conceptualized as objects, so that they can be relation-bearing items. It doesn t matter whether the reasons for this are conceptual, empirical, or something else entirely. Postulating objects is not an option for those structural realists who insist on the elimination of entities. But perhaps structural realists could insist that the usual mechanisms of detection ultimately allow us to detect properties and relations (presumably of the relevant objects). Access to detection properties (see Chakravartty 2007) canbeforgedby scientific instruments. And by combining access to such properties and the relevant relations, access to a particular structure emerges. In this way, it is specified what the structural realist is committed to. It is unclear, however, that this move will help structural realists, since the proper characterization of detection properties also ultimately presupposes objects the objects that have the relevant properties. As a result, structural realists would simply bebacktowheretheystarted. 6 (b) The commitment-to-objects argument also relies on the (widespread) use of set theory to characterize mathematical structures. Perhaps this argument as well as the Poincaré maneuver could be resisted by simply rejecting such use of set theory. We argue, below, that problems will emerge even if set-theoretic structures are not invoked. For the moment, note that the rejection of set theory comes with a significant cost for the structural realist. To begin with, recall that 6 More generally, one of the crucial features of Anjan Chakravartty s semirealism (Chakravartty 2007) is to argue that realists need the commitment to both objects and some properties and relations and, thus, some structures in order to get off the ground. Clearly, given the commitment to objects, this is not a move open to ontic structural realists.

10 an alleged virtue of the semantic approach is that it does not take one s theorizing about the sciences too far from actual scientific practice (as the syntactic approach arguably does; for an overview, see Suppe 2000). So, to avoid contradicting scientific practice and its widespread use of set theory, the structural realist who also adopts the semantic approach had better preserve the usual way set-theoretic structures are formulated and introduced in actual scientific practice. It would be disingenuous to dismiss the use of set-theoretic structures as irrelevant at this point. The way mathematicians and physicists introduce and formulate structures should be taken seriously in this context too. The result, however, is a commitment to objects as part of the resulting metaphysics. The structural realist may insist that set-theoretic structures only provide representational devices regarding the structures in question. One should not read off anything about the fundamental nature of the structures one should be committed to from the mere fact that they can be represented set-theoretically. If set-theoretic structures presuppose objects, so be it. This simply shows that these are not the structures the ontic structural realist is ultimately realist about. 7 A similar view is advanced by Brading and Landry in a series of papers (see Brading 2006, 2011 and Landry 2007). According to them, set theory plays no special role in characterizing structure and, in particular, in articulating the notion of shared structure, a central notion for any version of structuralism. Their suggestion is that this notion can be left unspecified (that is, it should not be assumed that it is a set-theoretic notion to begin with), and its nature should be decided on a case-by-case basis. All that matters is that we have a notion of shared structure. These responses, however, have a cost. Without the specification of the nature of the structures that the ontic structural realist is realist about, the very content of OSR is left unspecified. It then becomes unclear about what, exactly, the structural realist is realist. Without a clear characterization of the structures in question, the view ultimately lacks content. Thus, in order for OSR to get offthe ground,aproperspecification of structure is required. Furthermore, to advance, as Landry (2007) does, that the context determines the kind of characterization of structure required in each case falls prey to two difficulties. First, if the available options involve objects (as Landry seems to allow), then those who don t want to be committed to objects in the first place are not better off. Second, if the notion of structure is left unspecified, then one is left in the dark as to what one s realism is about. None of the options seem palatable to the OSR-theorist. But perhaps the structural realist could suggest that the specification of the relevant structures is done via ostension. Maybe there is no way of determining the scope of one s structural realism but by pointing to particular instances of the relevant structures about which one is a realist. The problem with ostension is that, for familiar Quinean reasons, it is radically indeterminate. One may point to an inscription on a piece of paper that represents, say, a set-theoretic structure, and state I m realist about that. But what does that refer to? The piece of paper? The inscription on the 7 This line of response has been suggested by Steven French and James Ladyman in conversation.

11 paper? The representation that is conveyed by the inscription? The object that is represented? The content of the representation? The physical interpretation associated with that content, and if so, which among the various such interpretations does one pick out? And how, exactly, can any such interpretation be picked out by ostension? Clearly ostension is entirely inadequate for the task at hand. One could try to avoid the commitment to objects by shifting from classical set theories to a non-classical set theory, such as quasi-set theory (for an exposition, see French and Krause 2006, Chapter7).Asiswellknown,quasi-settheoryallows for collections of things that lack identity conditions, the non-individuals. It is, thus, crucial for quasi-set theory that the extensionality axiom of classical set theories does not hold in general. After all, this axiom specifies identity conditions for every set, thus ruling out, by fiat, things that lack identity conditions: sets x and y are the same just in case they have the same members. The main motivation for introducing things that lack identity conditions is to model the behavior of non-individuals in quantum mechanics, according to the interpretation of the theory that admits of such things. Moreover, it is possible to define structure in quasi-set theory too, so that the elements of the domain could now be taken as being non-individuals. Given the restriction on the scope of the extensionality axiom, it may be thought that quasi-set theory could avoid the commitment to objects. Does that alleviate the burden on OSR? Not really. Even though some philosophers have advanced the idea that quantum mechanics with non-individuals is a version of OSR (in particular, see Votsis 2011),that isstill anobject-orientedontology. Non-individuals, as understood in quasi-set theory, are objects: one quantifies over them; they have certain properties (and lack others), and they bear relations to other things. As French (2010,p.94) makes clear, OSR does not get rid of the individuality of particular objects, it gets rid of objects altogether, whether they are individuals or not. This is important, since metaphysical underdetermination between the metaphysical packages of individuals and non-individuals is one of the main motivations for OSR. So, to adopt an alternative metaphysical package by allowing a set theory with non-individuals should not be seen as softening the burden for OSR. Non-individuals are objects too to take this path is ultimately to accept commitment to objects. 2.2 A plurality of structures Another significant difficulty for OSR, and for the Poincaré maneuver in particular, is that even if the latter managed to avoid commitment to objects in the characterization of set-theoretic structures, it is open to an important kind of underdetermination: it involves distinct but elementarily equivalent structures that are models of the same theory (Bueno 2011). 8 Due to the upward Löwenheim-Skolem theorem, first-order theories with models with infinite domains have elementarily equivalent but nonisomorphic models for every cardinality. The models are importantly different (since 8 Building from an argument advanced by Bueno 2011, thissectionexaminesadditionalconsiderations regarding the philosophical significance of elementarily equivalent but non-isomorphic models to the OSR debate.

12 they are non-isomorphic), but exactly the same first-order sentences are true in them (since they are elementarily equivalent). Which of those many models represents the structure of the world? That is, which of this huge number of structures is the structural realist realist about? An account of how one can choose among such structures and determine the right one needs to be offered. But it is unclear how this could be done. On what epistemic grounds can a structure be preferred over another that is elementarily equivalent to it? It seems that there is no simple, epistemic way to determine which particular structure is that of the world. Perhaps the choice among the various structures can be made based on pragmatic considerations, that is, considerations related to the users of the theory rather than based on epistemic, evidential grounds (see van Fraassen 1980). Pragmatic considerations include simplicity, familiarity, fecundity, and expressive power (the usual theoretical virtues). They provide reasons to prefer certain structures over others.it is undoubtedly easier to work with simpler, familiar structures, which are also fecund and have rich expressive power. However, this is a reason to accept the structures in question rather than believe that they properly describe the world (see van Fraassen 1980). After all, absent some metaphysical principle according to which the world itself is simple (in some sense), or that structures that are familiar, fecund, and rich in expressive power are more likely to describe reality than unfamiliar, barren, and inexpressive ones, pragmatic reasons alone are not sufficient to support the conclusion that the chosen structure is correct. Thus, a choice on purely pragmatic grounds is unable to support the realist component of the view. For if we were to choose pragmatically what the structure of the world is, we would not thereby have grounds to believe that such a structure is right. As a result, with multiple non-equivalent structures available, and no epistemic reason to choose between them, a case of underdetermination arises for the metaphysics of structures underlying OSR. In the end, it is unclear that the structural realist has the resources to specify the particular structure one should be realist about. But perhaps there is a way out here; one that is usually invoked in the defense of the superiority of the semantic approach over the syntactic view. Only the intended models of the theory in question are picked out. The fact that the semantic approach can accommodate this move is an important benefit of the view and a significant reason to prefer it over the syntactic approach (see Suppe 2000). However, this way out is not open to the structural realist. How is the choice of the intended model supposed to be made? Once again, to invoke pragmatic considerations as the basis to determine the nature of reality is not an available route. What is required is a structural, epistemic constraint on the choice of the structure of the world. But which structural, epistemic constraint could be invoked in the choice of the intended model? One would need to have independent reasons to believe that the fact that the intended model is intended somehow makes it more likely to be the right one the one that corresponds to the structure of the world. But no reason has been provided as to why this is the case. And it is unclear that there is such a reason available to the structural realist. It simply begs the question to assert that the intended model is natural, in the sense that a natural model provides the correct description of the structure of the world. Moreover, if by natural it is meant that the relevant models capture natural kinds, it is not obvious that such a move would be open to the structural realist

13 either. For the postulation of natural kinds introduces an ontology of objects those that have the relevant kinds and that is precisely what the ontic structural realist is trying to avoid. Alternatively, if kinds are identified extensionally, in terms of the sets of objects of the relevant kinds, the concerns raised earlier about the ontological commitment to sets which are ultimately objects, after all arise again. A further problem prompted by the existence of elementarily equivalent nonisomorphic models concerns the very idea of re-conceptualizationofobjects.recall that for the kind of OSR we are considering here, objects are derived from structures, they are either contextually individuated or merely the nodes in a web of relations. But even supposing that we could somehow fix a common underlying structure among those non-isomorphic models, there would be trouble with the number of objects that such a structure gives rise to. If we are going to take seriously the claim that objects are nodes in the web of relations or that they are individuated contextually by the relations of the structure, the cardinality of objects obtained in this way should be fixed. That is, one would expect that the structure of the world should give rise to one world, which has a well-determined number of objects (exactly the number of objects in reality), even if objects are to have only a secondary metaphysical status. However, due to the argument above, the same theory may give rise to structures with distinct domains, of distinct cardinalities. Using the vocabulary introduced above, reading a structure D, R from right to left may be performed in several distinct ways, each of them giving rise to a set D of distinct cardinality, and each of these sets could be the domain of a model of the theory and, thus, each could claim rights to be the one that properly represents reality. The structural realist may complain that to assume that there is a well-determined number of objects in the world is too stringent a requirement. It is not possible to determine that number without providing individuation conditions for objects. And due to vagueness, indeterminacy, or intractability, it may not be possible to determine what that number is. Let us grant this point. Despite that, presumably the structural realist who is willing to allow for a reconceptualization of objects in terms of structures also allows for there being some number of objects in the world. The determination of that number need not be made sharply. Perhaps the structural realist only indicates that the relevant number is within a certain range. However this determination is implemented, the problem just raised will arise again. For sets of distinct cardinality would emerge from reading the relevant structures from right to left, and each of these sets could be used as the domain of a model of the theory that represents the world as long as the cardinality of the domains is within the specified range. Alternatively, if no range at all is specified, then it becomes unclear why the structural realist intends to re-conceptualize objects in terms of structures. If there is no number of objects in the world, if not even a range for that number can be provided, the structural realist seems to lack a reasonable motivation to introduce such objects in the first place. Before we proceed, we should make it clear that the previous arguments are not a restatement of the well-known Newman objection presented to epistemic versions of structural realism. According to the Newman objection, attempts to articulate the theoretical content of a scientific theory (such as through its Ramsey sentence) fail to specify the precise extension of the theoretical relations. In fact, given any set with the

14 same cardinality as the intended model, we may convert that set into a model of the theory (see Ladyman 2013 for general discussion). 9 Our point, in contrast, focuses on the difficulties that non-isomorphic, but elementarily equivalent models which, thus, have distinct cardinalities raise to OSR; it goes in the opposite direction than Newman s. While Newman s objection moves from collections of objects with the same cardinality to relations, we go from relations to collections of objects with distinct cardinalities. Since the relationship between objects and relations in OSR is supposed to be such that the former are derived from the latter, our argument shows that such an operation, however implemented, can be executed in a plurality of ways. No structural constraint determines a particular domain as the correct one. As a result, this is not a version of Newman s objection. In Section 3,whenweexamine metaphysical characterizations of the relationship between structures and objects, we argue that additional difficulties emerge as well. But, once again, the argument proceeds from relations to objects, not from objects to relations. Perhaps that problem of the existence of multiple structures can be overcome if we use a higher-order logic. 10 With second-order logic we obtain categoricity for important mathematical theories, so that non-standard models are avoided in those cases. However, there is a price to be paid, and it isunclearthatthedesiredresultcanbe reached. First, as is well known, categoricity for higher-order theories only obtains when what is called standard semantics is taken into account, that is, a semantics in which the higher-order variables for properties and relations run through the whole plethora of properties and relations available. However, when Henkin semantics is employed, that is, the one in which variables run through some (but not necessarily all) subsets of the whole domain of relations and properties, non-standard models appear again, and a version of the Löwenheim-Skolem theorem holds. Even if we could reasonably choose only standard models (that is, models invoked in standard semantics) in a way acceptable to the structural realist, there would still be difficulties: (a) It is not clear that our best empirical theories are categorical, so the problem of determining what the right structure is would not be avoided. (b) Higher-order logics using standard models are incomplete. And it is unclear how structural realists can accommodate such incompleteness. Which status should they assign to statements that are true but not derivable from the relevant principles? (c) Objects are an integral part of the formalism of second-order logic, in the sense that any interpretation of such formalism whether in set theory or in some other formal framework presupposes objects. So, in the end, the OSR-theorist doesn t solve the problem by shifting to higher-order logics. A different proposal concerning the relation between objects and structures recommends the use of category theory instead of set theory (see Landry 2007 and Bain 2013). It is argued that category theory is better equipped to deal with the elimination of objects because categories are not defined in terms of objects, but rather in terms of morphisms (or arrows). There is no need to appeal to any kind of maneuver 9 Demopoulos (2003) alsodiscussesthisworry,andhelinksittoputnam smodel-theoreticargumentand to the semantic view of theories, but it is independent from the concerns we raise here. 10 For an excellent discussion of second-order logic, see Shapiro (1991).

15 here: objects are already given a secondary place. So it seems that category theory deals more adequately with the elimination of objects required by OSR and provides abetterrepresentationalsystemfortheview. One worry with this proposal is that the choice between set theory and category theory is being made on pragmatic grounds, given the expressive resources of category theory and those of set theory. But it is unclear why having certain expressive resources, such as being able to formulate structures without presupposing objects, is sufficient to ensure a realist reading of the categorial framework as the one that provides the proper characterization of the structure of the world. One would need to offer reasons as to why such a pragmatic choice will deliver structures that properly describe the world something that is needed given the intended realism about structure. However, in light of the considerations made above, it is not clear that pragmatic reasons, such as the expressive resources of the categorial approach, are good epistemic guides: they may provide reasons to accept the category-theoretic framework, but these need not be reasons to believe that the framework is true, or likely to be so (see van Fraassen 1980). The category theorist may respond by noting that the adoption of category theory is not done on pragmatic grounds: set theory is just inadequate to represent objectless structures, and so it fails to express properly what needs to be expressed. Category theory, in turn, is adequate to the task at hand. Thus, its adoption is not made on the basis of pragmatic considerations, but emerges from the adequacy of the expressive resources of the theory itself. However, is category theory really adequate in the relevant respect? We don t think it is. After all, the definition of a category presupposes objects. A category is defined in terms of objects and arrows (see Awodey 2010,pp. 4-5): For each arrow, there are objects, the domain and the codomain of the arrow. For each object there is an arrow (the identity arrow of that object). Given two arrows such that the codomain of one is identical to the domain of the other, there is an arrow which is their composite. The composition of arrows is required to be associative (that is, the composite of the composite of arrows f and g and the arrow h is identical to the composite of the arrow f with the composite of the arrows g and h as long as f s codomain is identical to g s domain, and g s codomain is identical to h s domain, so that the relevant compositions are defined). All arrows are required to have a unit (that is, for all arrows f,thecompositeof the identity arrow of f s domain and f is identical to the composite of f and the identity arrow of f s codomain, and both such compositions are identical to f ). Clearly, identity is presupposed throughout this definition: in particular, in the characterization of the composite arrow (whichpresupposestheidentityofthedomainof an arrow and the codomain of another), as well as in the formulation of associativity and unit (both of which presuppose the identity of the relevant arrows). Thus, genuine objects are presupposed: one quantifies over them, they have certain properties (e.g., each object has an identity arrow) and lack others (e.g., an object can be distinguished from an arrow), and they bear relations to other objects and arrows (some objects are domains of an arrow and codomains of another arrow, others are not).